Add AutoDream durable memory to builtin _memory plugin

README.md

@@ -73,6 +73,7 @@ docker run -p 50001:80 agent0ai/agent-zero
 
 - Agent Zero is not pre-programmed for specific tasks (but can be). It is meant to be a general-purpose personal assistant. Give it a task, and it will gather information, execute commands and code, cooperate with other agent instances, and do its best to accomplish it.
 - It has a persistent memory, allowing it to memorize previous solutions, code, facts, instructions, etc., to solve tasks faster and more reliably in the future.
+- Its AutoDream layer also maintains a Claude Code-style `MEMORY.md` index for each memory scope or project, backed by durable markdown memories and synchronized with the vector database so symbolic navigation and semantic recall stay aligned.
 
 ![Agent 0 Working](/docs/res/ui_screen2.png)
 

plugins/_memory/README.md

@@ -19,6 +19,30 @@ This plugin stores memories and knowledge embeddings in a FAISS-backed vector da
   - Exposes search, delete, bulk delete, update, and subdirectory listing endpoints for the memory dashboard.
 - **Scoped storage**
   - Supports different memory subdirectories so memory can be separated by context or agent scope.
+- **AutoDream durable memory**
+  - Periodically performs a reflective background pass over recent sessions and recent memory fragments.
+  - Maintains a compact `MEMORY.md` index plus durable markdown memory files under each memory subdirectory.
+  - Feeds the dream with both fresh vector memories and semantically related older vector memories so consolidation has better historical context.
+  - Writes a lightweight `.dream-log.md` changelog for fast auditing of what each run created, updated, pruned, or flagged.
+  - Re-imports durable memory files into vector memory with frontmatter metadata preserved, so recall benefits from the consolidated output without embedding raw YAML headers.
+
+## Claude Code `MEMORY.md` Integration
+
+Agent Zero's AutoDream intentionally follows the same broad pattern as Claude Code's `MEMORY.md`, but it does so per memory scope instead of assuming one global file.
+
+- **Scoped index, not a dump**
+  - Each memory scope gets its own `autodream/MEMORY.md`.
+  - That file is a compact index of durable memories, not the memories themselves.
+- **Durable files hold the actual content**
+  - The detailed memories live in `autodream/memories/*.md`.
+  - `MEMORY.md` links to those files with one-line descriptions.
+- **Vector DB stays in sync**
+  - Those durable markdown files are re-imported into the FAISS-backed vector store.
+  - This means Agent Zero's semantic recall and the Claude-style `MEMORY.md` index are two synchronized views over the same durable memory set.
+- **Compatibility model**
+  - If you already use Claude Code, the generated `MEMORY.md` should feel familiar.
+  - The main difference is placement: Agent Zero manages `MEMORY.md` inside each memory scope or project metadata folder rather than at a single repository root.
+  - If another tool expects a root-level `MEMORY.md`, you can mirror or sync the generated index there, but Agent Zero's native source of truth remains the scoped AutoDream folder.
 
 ## Key Files
 
@@ -28,6 +52,8 @@ This plugin stores memories and knowledge embeddings in a FAISS-backed vector da
   - `helpers/knowledge_import.py` imports external knowledge into memory storage.
 - **Consolidation**
   - `helpers/memory_consolidation.py` contains memory consolidation logic.
+- **AutoDream**
+  - `helpers/auto_dream.py` contains session gathering, durable file generation, and background scheduling logic.
 - **Tools**
   - `tools/memory_save.py`
   - `tools/memory_load.py`
@@ -37,6 +63,8 @@ This plugin stores memories and knowledge embeddings in a FAISS-backed vector da
 - **API**
   - `api/memory_dashboard.py` powers the memory management dashboard.
   - `api/import_knowledge.py` and `api/knowledge_reindex.py` handle knowledge import and reindexing.
+- **Lifecycle**
+  - `extensions/python/process_chain_end/_60_auto_dream.py` schedules periodic AutoDream runs after top-level chats complete.
 
 ## Configuration Scope
 

plugins/_memory/default_config.yaml

@@ -13,4 +13,8 @@ memory_recall_post_filter: false
 memory_memorize_enabled: true
 memory_memorize_consolidation: true
 memory_memorize_replace_threshold: 0.9
-agent_memory_subdir: default
+memory_auto_dream_enabled: false
+memory_auto_dream_min_hours: 8
+memory_auto_dream_min_sessions: 3
+memory_auto_dream_line_limit: 120
+agent_memory_subdir: default

plugins/_memory/extensions/python/process_chain_end/_60_auto_dream.py

@@ -0,0 +1,35 @@
+from agent import AgentContextType
+from helpers import persist_chat, plugins
+from helpers.extension import Extension
+
+from plugins._memory.helpers.auto_dream import schedule_auto_dream
+from plugins._memory.helpers.memory import get_agent_memory_subdir
+
+
+class AutoDream(Extension):
+
+    async def execute(self, **kwargs):
+        if not self.agent:
+            return
+        if self.agent.number != 0:
+            return
+        if self.agent.context.type == AgentContextType.BACKGROUND:
+            return
+
+        config = plugins.get_plugin_config("_memory", self.agent) or {}
+        if not config.get("memory_memorize_enabled"):
+            return
+        if not config.get("memory_auto_dream_enabled"):
+            return
+
+        if self.agent.config.profile:
+            self.agent.context.set_data("agent_profile", self.agent.config.profile)
+
+        persist_chat.save_tmp_chat(self.agent.context)
+
+        schedule_auto_dream(
+            context_id=self.agent.context.id,
+            project_name=self.agent.context.get_data("project"),
+            agent_profile=self.agent.config.profile,
+            memory_subdir=get_agent_memory_subdir(self.agent),
+        )